Abstract Data Machine
Muhammad Taimoor Khan, Anastasios Fragopoulos, Dimitrios N. Serpanos, Howard E. Shrobe · 2015
In this paper, we present our ongoing work and formalism of a novel data classification method (Abstract Data Machine) for reliable software systems. Most of the approaches for data classification are based on statistical classification, e.g. machine-learning algorithms that comes with false rates. Another critical problem with such algorithms is that they are not reliable and thus do not ensure any provable assurances. One approach to establish reliability of the algorithms is to classify arbitrary data with an appropriate (pragmatically useful) level of abstraction based on the logical data properties that are amenable to assurances, i.e. a formal proof that data represented by a class indeed respects properties of the class. Furthermore, the formal proofs are fundamental for the security assurance of the applications using the algorithms in general and data security in particular.